Consumer Adoption of Autonomous Robots for Last-Mile Deliveries

Final Analysis

Author

Adwoa Darko , Siddharth Kusuma, Sudeep Januga

Published

December 12, 2025

Abstract

This study investigates consumer adoption of autonomous delivery technologies—specifically robots and drones for last-mile delivery. The project explores how customers trade off between key delivery attributes: price, speed, environmental sustainability (CO₂ emissions), reliability, and delivery mode (Truck, Robot, Drone). A pilot conjoint survey was conducted among frequent e-commerce shoppers who regularly order small and light packages (under 10 lbs).

The results of the conjoint analysis demonstrate that participants made their delivery choices primarily based on price, delivery speed, and environmental impact. Among these, price emerged as the strongest and most statistically significant factor. Consumers showed a clear tendency to avoid higher-priced delivery options, even if other features were attractive, reflecting strong cost sensitivity. Speed was the next most influential factor, with faster deliveries significantly increasing the probability of selection—highlighting the growing consumer expectation for near-instant fulfillment. Emissions also played an important role: lower-carbon delivery options were consistently favored, indicating that environmental consciousness is now part of real consumer decision-making rather than a purely moral statement.

By contrast, reliability (within 85–95%) and delivery mode (Truck, Robot, Drone) did not significantly affect choice behavior. This suggests that participants perceive all delivery options as sufficiently reliable and are largely indifferent to whether a human, robot, or drone delivers the package—as long as the service is fast, affordable, and environmentally friendly. In practice, this implies that companies introducing automated delivery technologies should focus their marketing on tangible consumer benefits like cost savings, speed, and sustainability rather than the novelty of the technology itself. The neutrality toward robots and drones indicates readiness for adoption—autonomous delivery is no longer resisted, but to gain market traction, it must deliver clear value in convenience and price.

Introduction

The rapid rise of e-commerce has led to an exponential increase in small-package deliveries, making the last-mile segment both critical and costly for logistics companies. Autonomous delivery technologies such as ground robots and drones are being developed to make deliveries faster, cheaper, and more sustainable, but public acceptance remains uncertain. Understanding how consumers value attributes like delivery cost, time, and sustainability is essential before large-scale deployment.

This study focuses on evaluating consumer preferences for different delivery modes (Truck, Robot, Drone) and the trade-offs among four major attributes: Price, Speed, CO₂ Emissions, and Reliability. Each attribute was varied at realistic market levels to simulate decision-making scenarios. For example, prices ranged from $2 to $8, speeds from 30 minutes to 24 hours, emissions from 2 lb to 0.001 lb of CO₂ per mile, and reliability from 85% to 95%. A choice-based conjoint (CBC) survey was designed to estimate the relative importance of each attribute.

The analysis contributes to the broader policy discussion on the sustainability and feasibility of autonomous last-mile systems, offering insights into consumer willingness to adopt technology-driven delivery methods that could reduce emissions and traffic congestion.

Survey Design

Eligibility and Target Population

The target population comprised frequent online shoppers who regularly receive small or light packages (under 10 lbs). Eligibility criteria ensured relevance to real-world robot delivery applications:

Must order from e-commerce retailers such as Amazon, Walmart, or Target.
Receive at least one package per month.
Typically receive small or light items such as books, electronics, or clothing.
Respondents who only receive large or heavy packages were excluded

Data Collected

The survey collected the following key information:

Demographics: age, gender, education level, household income, and residential area.
Behavioral variables: online shopping frequency, number of monthly packages, delivery urgency, and typical spending on delivery fees.
Attitudinal measures: environmental concern, trust in autonomous delivery, and comfort level with robots or drones

Educational Material

Before choice questions, respondents viewed a brief educational page explaining each delivery attribute in plain language:

Price: delivery fee in dollars.
Speed: delivery time (e.g., 30 minutes, 1 hour, or 12–24 hours).
Sustainability (CO₂ emissions): pounds of CO₂ per mile emitted during delivery, with examples comparing truck and drone emissions.
Reliability & Safety: the likelihood that a package arrives on time and undamaged

This step ensured all respondents understood trade-offs before making their choices.

Conjoint Design

Each respondent answered six choice-based conjoint questions, each presenting three delivery options with randomized attribute combinations. No “none” option was included, compelling a realistic trade-off.

Attributes and Levels:

Attribute	Levels	Unit
Price	2, 5, 8	$ (dollars)
Speed	12-24 hrs, 60 min, 30 min	Time in min & hrs
Sustainability	2, 0.05, 0.001	Pounds of CO₂ per mile
Reliability & Safety	85%, 90%, 95%	% successful delivery

Example Question

Data Analysis

Sample Description

The survey sample includes 165 unique respondents who passed the screening criteria and consented, contributing 990 completed choice sets. These respondents also provided answers to demographic and background questions used to describe the sample and explore heterogeneity in delivery preferences.

Respondents range from early 20s to over 70 years old, with the age distribution centered in the late 30s to mid‑40s, as shown in the age histogram. Gender is approximately balanced, with slightly more women than men and a very small number of non‑binary respondents.

The sample is predominantly white, with smaller shares of Black, Hispanic, Asian, and Native American respondents, as illustrated in the race/ethnicity bar chart. Educational attainment is relatively high overall: most respondents report at least some college, and bachelor’s or associate degrees are the most common categories, with smaller groups having high school only or graduate degrees.

Household income is concentrated in the middle ranges, with many respondents between 40,000 and 149,000 USD per year and fewer at the very low or very high ends of the distribution. Package delivery is already a routine activity for this group: most respondents receive packages once a week or several times per week, and only a small minority report rarely receiving packages

Data Cleaning

We began with 215 total survey attempts in the final deployment. The first quality step was to remove respondents who failed the built-in eligibility screener. In this survey, individuals were screened out if they did not meet the study’s basic criteria—for example, not being an active e-commerce shopper or indicating that they do not order small/light packages. After filtering out screened respondents, 204 responses remained. This step ensures that the analysis reflects the intended target population of online shoppers whose delivery preferences are meaningful for the study.

Next, we restricted the dataset to respondents who completed all six choice-based conjoint (CBC) tasks. Anyone missing one or more CBC questions was removed, reducing the dataset from 204 to 201 respondents. We then applied a standard “flatliner” check to identify respondents who selected the exact same alternative in every choice task, a behavior that typically signals low engagement or satisficing rather than thoughtful trade-off evaluation. Respondents exhibiting this pattern were removed, leaving 198 valid respondents after this step.

Finally, we screened for unrealistic completion times, which help identify respondents who may have rushed through the survey without adequately processing the attribute information. We converted total response time into minutes and examined the distribution. Based on this, we excluded respondents who completed the choice-task portion in under 5 minutes, a threshold indicating insufficient cognitive effort for evaluating six multi-attribute profiles. After applying the time-quality filter, the final analytic sample consisted of 165 high-quality, engaged respondents. Thus, out of the original 215 survey attempts, 50 respondents were removed across screening, completeness, behavioral, and speed checks, ensuring that the remaining dataset was reliable, attentive, and appropriate for estimating the conjoint preference models.

Modeling

Simple Logit Model

We estimated a Simple Logit model to quantify how each delivery attribute influenced respondents’ choices in the conjoint experiment.

\[ u_j = \beta_1 x_j^{\mathrm{price}} + \beta_2 x_j^{\mathrm{speed}} + \beta_3 x_j^{\mathrm{emissions}} + \beta_4 x_j^{\mathrm{reliability}} + \beta_5 \delta_j^{\mathrm{Robot}} + \beta_6 \delta_j^{\mathrm{Drone}} \] The results for the logit model are summarized below:

parameter	Estimate	Std_Error	Z_value	P_value	Sig
price	-0.3564	0.0217	-16.39	<0.001	***
speed	-0.0349	0.0045	-7.82	<0.001	***
emissions	-0.5792	0.0541	-10.70	<0.001	***
reliability	0.1069	0.0119	9.00	<0.001	***
delivery_modeRobot	-0.0053	0.1123	-0.05	0.9620
delivery_modeDrone	0.0765	0.1131	0.68	0.4988

The multinomial logit results indicate that several delivery attributes strongly influence consumer choice, while others have minimal impact. Price, emissions, and reliability emerge as the most important drivers of preference. Higher prices and higher CO₂ emissions significantly reduce the likelihood that a delivery option is chosen, with emissions showing one of the strongest negative effects, suggesting that respondents place meaningful value on environmentally friendly delivery methods. Reliability has a positive and highly significant effect, indicating that consumers prefer options they can depend on. Speed also matters, with slower delivery times reducing choice probability, though its effect is smaller than that of price or emissions. In contrast, the delivery mode itself—whether a package is delivered by a robot or drone—shows no statistically significant impact, meaning respondents do not strongly prefer or avoid these newer delivery technologies relative to traditional truck delivery. Overall, consumers prioritize cost, environmental impact, and performance-related attributes over the specific technology used to deliver their packages.

Mixed Logit

parameter	Estimate	Std_Error	Z_value	P_value	Sig
price	-0.5109	0.0696	-7.34	<0.001	***
speed	-0.0559	0.0120	-4.65	<0.001	***
emissions	-0.8843	0.1549	-5.71	<0.001	***
reliability	0.1552	0.0277	5.61	<0.001	***
delivery_mode_Drone	0.2141	0.2443	0.88	0.381
delivery_mode_Robot	-0.2531	0.1925	-1.31	0.189
sd_speed	0.0698	0.0270	2.58	0.009	**
sd_emissions	0.9519	0.2775	3.43	0.0006	***
sd_reliability	-0.1534	0.0692	-2.22	0.027	*
sd_delivery_mode_Drone	0.3441	0.9495	0.36	0.717
sd_delivery_mode_Robot	-1.4459	0.5981	-2.42	0.016	*

The mixed logit results show strong and consistent preferences across respondents: people dislike higher prices, slower delivery, and higher emissions, while valuing greater reliability. Unlike the simple MNL model, the MXL reveals substantial heterogeneity—speed, emissions, reliability, and delivery mode all have significant random-parameter variation, meaning different individuals value these attributes to very different degrees. Delivery mode (Drone or Robot) does not have a significant average effect overall, but the significant variance for the Robot option indicates polarized opinions: some respondents prefer it while others avoid it. The WTP-space model confirms these findings in dollar terms, showing meaningful willingness to pay for improved reliability and reduced emissions. Overall, the mixed logit model provides a better, more realistic representation of individual-level differences in preferences.

Subgroup Logit Model

Attribute	Group A Mean	A 95% CI Lower	A 95% CI Upper	Group B Mean	B 95% CI Lower	B 95% CI Upper
speed	-0.098	-0.137	-0.061	-0.096	-0.132	-0.063
emissions	-1.147	-1.601	-0.730	-1.932	-2.394	-1.524
reliability	0.214	0.117	0.316	0.357	0.265	0.459
delivery_mode_Drone	0.265	-0.657	1.179	0.209	-0.654	1.072
delivery_mode_Robot	0.199	-0.723	1.098	-0.106	-0.939	0.751

The WTP comparison across income groups reveals meaningful differences in how lower-income (<$50k) and higher-income (≥$50k) respondents value key delivery attributes. Both groups show a positive willingness to pay for enhanced drone delivery and greater reliability, but the effect is notably stronger among higher-income respondents, who exhibit nearly double the WTP for reliability improvements. Emissions reductions also matter to both segments, yet higher-income respondents display a substantially larger negative WTP for higher emissions, indicating a stronger preference for environmentally friendly delivery options. In contrast, speed and robot delivery attributes generate relatively small WTP values in both groups, suggesting these features are less influential in driving choices. Overall, the results indicate that higher-income consumers are more sensitive to environmental quality and service reliability, while lower-income consumers demonstrate more moderate but still positive valuations for premium delivery features.

Results

WTP

Attribute	Mean WTP ($)	Lower 95% CI	Upper 95% CI
Speed	-0.098	-0.124	-0.073
Emissions	-1.628	-1.955	-1.323
Reliability	0.300	0.233	0.371
Drone delivery	0.220	-0.404	0.844
Robot delivery	-0.012	-0.619	0.607

The willingness-to-pay (WTP) results reveal clear patterns in the attributes that most strongly influence consumer delivery preferences. Across all plots, increases in delivery speed (longer wait times) and emissions consistently reduce WTP, indicating that consumers prefer faster and environmentally cleaner delivery options and are willing to pay to avoid undesirable levels of each. Reliability emerges as one of the most influential attributes: as reliability increases from 85% to 95%, WTP rises steeply, with wide confidence intervals reflecting substantial heterogeneity in how much consumers value dependable service. In contrast, preferences for delivery mode—drone or robot relative to truck—show small mean WTP values and wide 95% confidence intervals that overlap zero, suggesting that consumers do not strongly prefer one mode over another given current perceptions. Together, these results indicate that reliability, emissions, and delivery speed are the key drivers of consumer choice, while delivery mode exerts comparatively weaker influence.

Market Simulation

To evaluate how key delivery attributes influence consumer choice, a set of structured market simulation scenarios was created. Each scenario isolates one attribute—price, emissions, delivery mode, reliability, or speed—while holding the remaining attributes constant. This allows for a clear understanding of how each attribute independently shifts market share among competing alternatives.

The baseline attribute levels reflect realistic ranges used in the conjoint survey:

Price: $2, $5, $8
Speed: 24 hr, 1 hr, 0.5 hr
Emissions: 2 lb, 0.05 lb, 0.001 lb CO₂
Reliability: 85%, 90%, 95%
Delivery Mode: Truck, Drone, Robot

These values were selected because they span low, medium, and high performance levels while remaining consistent with the respondents’ choice tasks. The simulations help illustrate competitive positioning when one attribute improves or worsens.

Key Insights from Market Simulations

Price is the strongest driver of market share. Reducing price leads to the largest increases in predicted adoption, making it the most influential attribute in shaping consumer preference.
Reliability is the most impactful non-price attribute. Moving from low to high reliability produces a substantial jump in market share, indicating that consumers heavily value trustworthy and consistent delivery performance.
Speed improvements matter, but only up to a point. Reducing delivery time from 24 hours to 1 hour significantly boosts preference, but the additional improvement from 1 hour to 0.5 hours provides only a small incremental gain.
Emissions reductions are valued, but less than reliability or price. Lowering emissions increases share meaningfully, though the effect is weaker compared to high reliability or lower price.
Changing delivery mode alone (Drone, Robot, Truck) does not materially shift consumer preference. When all other attributes are held constant, consumers show similar market shares across modes, suggesting that performance attributes drive choice more than delivery method.
Overall market adoption is maximized by combinations of:
- Lower price
- Higher reliability
- Faster delivery (major improvements only)
- Lower emissions

These findings highlight that improving reliability and optimizing price strategy offer the greatest opportunities to increase consumer demand, with speed and emissions offering secondary advantages.

Sensitivity Analysis

Tornado Plot – Sensitivity to Other Attributes

The tornado plot summarizes how the drone’s market share responds to ±20% changes in each attribute relative to the baseline (price = $5, speed = 1 hr, emissions = 0.05 kg CO₂, reliability = 90%).
Reliability is by far the most influential non-price attribute:
- Increasing reliability from 90% to 108% (a stylized “very high reliability” case) increases predicted share from about 0.56 to 0.90.
- Decreasing reliability to 72% collapses share to roughly 0.16.
- This very wide bar dominates the tornado chart, indicating that reliability is the primary design lever beyond price.
Price has the next-largest impact:
- Lowering price by 20% (to $4) raises share to about 0.64, while increasing price by 20% (to $6) reduces share to about 0.47.
- This confirms the strong price sensitivity already visible in the price–share plot.
Speed (1 hr → 0.8 hr or 1.2 hr) and emissions (0.05 → 0.04 or 0.06 kg CO₂) have very small effects on market share in comparison:
- Predicted shares remain clustered around the baseline (≈ 0.56), with only minor deviations, indicating that small ±20% changes in these attributes do not dramatically shift demand in this configuration.

Design & Pricing Recommendations (with Uncertainty)

Best price point (revenue-maximizing):
- The revenue curve peaks around $5–$6, with overlapping 95% CIs, so this range should be treated as the optimal pricing band rather than a single exact value.
- Pricing significantly below $5 sacrifices revenue (too much margin lost), while pricing above $6 rapidly erodes both share and revenue.
Key design decisions to maximize demand:
- Prioritize reliability: moving toward very high reliability levels produces the largest gains in market share and should be a central engineering and operations target.
- Use price strategically: modest under-pricing (e.g., closer to $5 than $6) can secure higher share without sacrificing too much revenue.
- Maintain reasonable speed and low emissions, but recognize that, within the tested ranges, incremental improvements in these attributes are less impactful than reliability and price.
Overall, the sensitivity analysis indicates that reliability and price are the dominant drivers of market demand, while speed and emissions play supporting roles. The uncertainty bands around all plots remind us that these estimates are not exact; instead, they define plausible ranges within which the true optimal price and attribute levels likely lie.

Final Recommendations and Conclusions

Overall Competitiveness of the Product

The analysis indicates that the proposed delivery product—particularly the drone-based alternative—has strong potential to be competitive in the current market landscape. Across all simulations, drone delivery consistently achieved the highest predicted market share, outperforming both robot and truck delivery modes. The key drivers for its competitiveness are its low emissions profile and high reliability, which were among the top-valued attributes in the WTP model. The product is therefore well positioned for adoption, provided that pricing remains in a consumer-acceptable range and reliability targets are met.

Recommended Price Range and Revenue Implications

Sensitivity analysis of price demonstrated that market share declines smoothly as price increases, with revenue peaking at approximately $5–6, where both adoption and total revenue are maximized. At prices below $4, revenue falls due to margin constraints, while prices above $7 lead to steep declines in market share. Based on the revenue curve and its associated confidence intervals, the optimal price point is $5, with reasonable robustness across the confidence band. Estimates remain consistent across simulations, giving moderate confidence in the recommendation, though real-world factors such as retailer partnerships, promotional pricing, or operational costs could shift the optimal point slightly.

Key Uncertainties Affecting Profitability

Despite strong model fit, several uncertainties may influence profitability and competitive success, including:

Regulatory approval and operational feasibility for drone or robot delivery, which could delay or limit deployment in certain regions.
Consumer trust and safety perceptions, which are not captured fully in the choice experiment but could shape adoption significantly.
Competitor innovations, such as ultra-fast truck deliveries or new subscription bundles, which could alter price sensitivity.
External cost drivers, such as energy prices, congestion charges, or labor shortages, that may shift the economics of last-mile delivery.

These uncertainties imply that price and design recommendations should be updated as technology matures and more real-world performance data becomes available.

Recommended Actions for Product Design

Based on the WTP model, tornado sensitivity analysis, and market simulations:

Prioritize High Reliability: Reliability showed one of the strongest effects on market share. Increasing reliability from 90% to 108% resulted in a significant market share jump (to ~90%). Even moderate improvements yield noticeable gains.
Maintain Competitive Emissions Performance: While emissions improvements had slightly smaller marginal effects than reliability, lower emissions levels consistently produced higher adoption. This reinforces sustainability as a market advantage.
Avoid Extremely Fast Delivery Speeds Unless Operationally Efficient: Speed improves market share, but its effect is smaller relative to price and reliability. Consumers value speed, but not enough to justify major cost trade-offs.
Drone Mode as the Default Offering: Drone delivery offers the strongest baseline utility and the highest simulated market share. It should be positioned as the flagship offering, with robot delivery reserved for specific contexts (dense urban areas, sidewalks, or building interiors).

Opportunities for Increasing Demand

From the combined analyses, three major opportunities emerge:

Compete Strongly on Reliability: Reliability is the single most influential attribute in driving preference. Investments in navigation precision, weather handling, and delivery success rates will significantly increase market share and willingness to pay.
Optimize Pricing for Volume and Revenue: A price point around $5 balances adoption with strong revenue performance. This price also remains competitive relative to truck-based delivery while differentiating through sustainability and speed.
Leverage Sustainability Messaging: Emissions reductions meaningfully increase WTP—more than speed improvements. Positioning the service as a low-carbon delivery option could expand appeal, especially among environmentally conscious consumers.

Conclusion

The combined conjoint results, market simulations, and sensitivity analyses show that a competitively priced, highly reliable, low-emissions drone delivery service has strong market potential. The optimal price range centers around $5–6, with reliability improvements offering the greatest leverage for increasing adoption. While uncertainties remain—particularly in regulation, technology maturity, and consumer trust—the model provides robust evidence that the product can achieve meaningful market share and profitability under realistic conditions. Continuous validation with larger and more representative samples is recommended as the product moves closer to implementation.

Limitations

Sample Representatives: Although the final data set contains a sufficient number of completed responses, the sample is not fully representative of the broader population of e-commerce consumers. The survey disproportionately includes younger, student-aged respondents with similar income levels and purchasing patterns, which may not reflect the preferences of high-frequency online shoppers, older users, or individuals in regions with different delivery expectations and infrastructure.
Hypothetical Choice Behavior: Conjoint analysis relies on respondents making choices in a hypothetical environment. While this method reveals trade-offs, real-world behaviors may differ due to brand familiarity, trust in delivery providers, or risk perceptions—factors that were intentionally held constant in this study. Attributes such as drone or robot delivery may also trigger concerns (noise, safety, privacy) that were not measured but could affect true adoption.
Simplified Market Context: The model assumes a market with only three delivery alternatives and holds many contextual variables constant (e.g., delivery fees set by retailers, service availability by region, or time-of-day delivery variations). In reality, consumers face a wider competitive landscape, including same-day premium services, subscription bundles (e.g., Amazon Prime), and retailer-specific policies.
Static and Short-Term Perspective: The study captures preferences at a single point in time. Preferences for sustainability attributes, willingness to pay for speed, or comfort with autonomous systems may evolve as technology matures, regulations change, or users gain more exposure to autonomous delivery options.

Appendix

Welcome to the Survey

Thank you for taking part in our survey.

We are graduate students at The George Washington University conducting a study on “Consumer Adoption of Autonomous Robots for Last-Mile Deliveries”

Your responses are anonymous and will only be used for academic research purposes. There are no right or wrong answers—we are interested in your honest opinions.

Click Next to continue.

Code

sd_next()

Consent

This survey is being conducted by students at The George Washing University. We will not be collecting any identifying data, such as your name or address. The whole survey will take approximately 5 to 10 minutes to complete. Your participation is voluntary and you may stop the survey at any time.

If you would like to participate, please answer the following questions:

Code

sd_question(
  type  = 'mc',
  id    = 'consent_age',
  label = "I am age 18 or older",
  option = c(
    'Yes' = 'yes',
    'No'  = 'no'
  )
)

I am age 18 or older

Yes

Code

sd_question(
  type  = 'mc',
  id    = 'consent_understand',
  label = "I have read and understand the above information",
  option = c(
    'Yes' = 'yes',
    'No'  = 'no'
  )
)

I have read and understand the above information

Yes

Code

sd_next()

Order Packages

Code

sd_question(
  type  = 'mc',
  id    = 'order_packages',
  label = "Do you order packages from retailers like Amazon, Walmart etc:",
  option = c(
    'Yes' = 'yes',
    'No'  = 'no'
)
)

Do you order packages from retailers like Amazon, Walmart etc:

Yes

Code

sd_next()

Code

sd_question(
  type   = 'mc_buttons',
  id     = 'frequency',
  label  = "How often do you shop online at retailers like Amazon, Walmart, or Target?",
  option = c(
    "Less than Monthly" = "less_monthly",
    "Monthly once or Twice" = "monthly",
    "Weekly"  = "weekly",
    "Multiple times per week" = "often"
  ), 
  direction = "vertical"
)

How often do you shop online at retailers like Amazon, Walmart, or Target?

Code

sd_question(
  type = "slider_numeric", 
  id = 'slider_single_val',  
  label = "In a typical month, how many packages (excluding groceries or meals) do you receive
from online shopping?", 
  option = seq(0, 10, 1)
)

In a typical month, how many packages (excluding groceries or meals) do you receive from online shopping?

Code

sd_next()

Few more Questions

Code

sd_question(
  type  = 'mc',
  id    = 'screenout',
  label = "Do you order small packages regularly (under 10 lbs)?",
  option = c(
    'Yes'  = 'yes',
    'No' = 'no'
  )
)

Do you order small packages regularly (under 10 lbs)?

Yes

Code

sd_question(
  type  = "matrix",
  id    = "attitudes_delivery",
  label = "Please indicate how much you agree or disagree with each statement:",
  row  = c(
    "I typically need my online orders to arrive within 1 day." = "need_1day",
    "Environmental impact is an important factor in my delivery choice." = "env_importance",
    "I would feel comfortable receiving a package delivered by an autonomous robot." = "robot_comfort"
  ),
  option = c(
    "Disagree" = "d",
    "Neutral" = "n",
    "Agree"   = "a"
    
  )
)

Please indicate how much you agree or disagree with each statement:

	Disagree	Neutral	Agree
I typically need my online orders to arrive within 1 day.	Disagree Neutral Agree *
Environmental impact is an important factor in my delivery choice.	Disagree Neutral Agree *
I would feel comfortable receiving a package delivered by an autonomous robot.	Disagree Neutral Agree *

Code

sd_next()

Why We’re Doing This Survey

Great work! You are eligible to take our survey.

Online shopping has become part of everyday life, and millions of small packages are delivered across cities every day. With so many deliveries happening, companies are now exploring new ways to deliver packages faster, cheaper, and more sustainably — including the use of autonomous delivery robots and drones.

These new technologies could change how we receive our orders, but they also raise important questions:

What do customers value most — cost, speed of delivery, sustainability or reliability?
Would people be comfortable trusting a robot or drone to bring their package?
How can these services be designed to meet real customer needs?

The purpose of this survey is to understand how consumers like you make decisions about different delivery options. Your responses will help researchers identify which delivery features matter most and guide how future autonomous delivery systems are developed and introduced.

Code

sd_next()

Educational Page

Now we’d like you to consider a scenario in which your packages are being delivered by different modes i.e either by Truck, Robot, Drone, so you can choose which one do you prefer by comparing their attributes.

Before we ask you any questions, let’s learn a little bit more about each of these attributes:

Price:

The delivery fee you would pay for a package.

Speed:

How quickly your package arrives (examples: within 30 minutes, 1 hour, or 12–24 hours).

Sustainability (CO₂ emissions):

The amount of carbon dioxide (CO₂) released per mile during delivery. For reference, a typical delivery truck emits 1 lb of CO₂ per mile.

A drone would have to fly 1,000 miles to emit the same amount of CO₂ as a truck driving just one mile.

Reliability & Safety:

The chance your package arrives on time, undamaged, and secure.

Delivery Modes

We will show 3 different types of delivery modes to choose from:

Truck	Robot	Drone

Code

sd_next()

Choice-based Question Practice

We’ll now begin the choice tasks. On the next few pages we will show you three options of delivery modes and we’ll ask you to choose the one you most prefer by comparing all attributes.

Code

# Define the option vector

package_buttons_option <- c("option_1", "option_2", "option_3")

# Change the names of each element to display markdown-formatted
# text and an embedded image using html

names(package_buttons_option)[1] <- "
  **Option 1**<br>
  <img src='images/truck.png' width=150><br>
        **Delivery Mode**: Truck<br>
        **Price**: $ 2<br>
        **Speed**: 24 hr<br>
        **Emissions**: 2 lb of CO₂<br>
        **Reliability**: 85 %
"
names(package_buttons_option)[2] <- "
  **Option 2**<br>
  <img src='images/robot.png' width=150><br>
       **Delivery Mode**: Robot<br>
        **Price**: $ 5<br>
        **Speed**: 1 hr<br>
        **Emissions**: 0.05 lb of CO₂<br>
        **Reliability**: 90 %
"
names(package_buttons_option)[3] <- "
  **Option 3**<br>
  <img src='images/drone.png' width=150><br>
        **Delivery Mode**: Drone<br>
        **Price**: $ 8<br>
        **Speed**: 0.5 hr<br>
        **Emissions**: 0.001 lb of CO₂<br>
        **Reliability**: 95 %
"

sd_question(
  type   = 'mc_buttons',
  id     = 'cbc_practice',
  label  = "Which delivery option would you choose among these?",
  option = package_buttons_option,
    
)

Which delivery option would you choose among these?

Code

sd_next()

Conjoint Question Intro

Great work!

We will now show you 6 sets of choice questions starting on the next page.

Code

sd_next()

Question 1

Code

sd_output("cbc_q1", type = "question")

Code

sd_next()

Question 2

Code

sd_output("cbc_q2", type = "question")

Code

sd_next()

Question 3

Code

sd_output("cbc_q3", type = "question")

Code

sd_next()

Question 4

Code

sd_output("cbc_q4", type = "question")

Code

sd_next()

Question 5

Code

sd_output("cbc_q5", type = "question")

Code

sd_next()

Question 6

Code

sd_output("cbc_q6", type = "question")

Code

sd_next()

Demographics Page

Nice job!

We’re almost done! We’d just like to ask just a few more questions about you which we will only use for analyzing our survey data.

Code

years <- as.character(2003:1920)
names(years) <- years
years <- c("Prefer not to say" = "prefer_not_say", years)

sd_question(
  type   = 'select',
  id     = 'year_of_birth',
  label  = "(1) In what year were you born?",
  option = years
)

(1) In what year were you born?

Code

genders <- c(
  "Male"                     = "male",
  "Female"                   = "female", 
  "Non-binary"               = "non_binary", 
  "Trans male/trans man"     = "trans_male",
  "Trans female/trans woman" = "trans_female",
  "Prefer not to say"        = "prefer_not_to_say"
)

sd_question(
  type   = 'select',
  id     = 'gender',
  label  = "(2) What is your current gender identity?",
  option = genders
)

(2) What is your current gender identity?

Code

races <- c(
  "White (Not of Hispanic or Latino origin)"  = "white",
  "African American or Black"                 = "black",
  "Asian"                                     = "asian",
  "Hispanic or Latino"                        = "hispanic",
  "American Indian or Alaska Native"          = "native_american",
  "Native Hawaiian or Other Pacific Islander" = "pacific_islander",
  "Prefer not to say"                         = "prefer_not_to_say"
)

sd_question(
  type   = 'select',
  id     = 'race',
  label  = "(3) I identify my race as (select all that apply):",
  option = races
)

(3) I identify my race as (select all that apply):

Code

educations <- c(
  "Less than high school" = "less_than_high_school",
  "High school"           = "high_school",
  "Some college"          = "some_college",
  "Associate's degree"    = "associate_degree",
  "Bachelor's degree"     = "bachelor_degree",
  "Master's degree"       = "master_degree",
  "Doctoral degree"       = "doctoral_degree",
  "Prefer not to say"     = "prefer_not_to_say"
)

sd_question(
  type   = 'select',
  id     = 'education',
  label  = "(4) What is the highest degree or level of school you have completed? If currently enrolled, please use the highest degree received.",
  option = educations
)

(4) What is the highest degree or level of school you have completed? If currently enrolled, please use the highest degree received.

Code

incomes <- c(
  "Less than $10,000"    = "less_than_10k",
  "$10,000 to $14,999"   = "10k_to_14k",
  "$15,000 to $19,999"   = "15k_to_19k",
  "$20,000 to $24,999"   = "20k_to_24k",
  "$25,000 to $29,999"   = "25k_to_29k",
  "$30,000 to $34,999"   = "30k_to_34k",
  "$35,000 to $39,999"   = "35k_to_39k",
  "$40,000 to $49,999"   = "40k_to_49k",
  "$50,000 to $74,999"   = "50k_to_74k",
  "$75,000 to $99,999"   = "75k_to_99k",
  "$100,000 to $149,999" = "100k_to_149k",
  "$150,000 to $199,999" = "150k_to_199k",
  "$200,000 or more"     = "200k_or_more",
  "Prefer not to say"    = "prefer_not_to_say"
)

sd_question(
  type   = 'select',
  id     = 'income',
  label  = "(5) What is your annual household income (from all sources) before taxes and other deductions from pay?",
  option = incomes
)

(5) What is your annual household income (from all sources) before taxes and other deductions from pay?

Code

sd_question(
  type  = "textarea",
  id    = "feedback",
  label = 
  "Please let us know if you have any other thoughts or feedback on this survey.
  
  Your feedback will help us make future improvements :)"
)

Please let us know if you have any other thoughts or feedback on this survey.

Your feedback will help us make future improvements :)

Code

sd_next()

End Page

The survey is now finished. You may close the window.

Code

sd_close()

--- title: "Consumer Adoption of Autonomous Robots for Last-Mile Deliveries" subtitle: "Final Analysis" author: Adwoa Darko , Siddharth Kusuma, Sudeep Januga date: December 12, 2025 format: html: toc: true toc-location: right theme: united code-fold: true code-tools: true self-contained: true --- ```{r} #| label: setup #| include: false #| echo: false knitr::opts_chunk$set( warning = FALSE, message = FALSE, fig.path = "figs/", fig.width = 7.252, fig.height = 4, comment = "#>", fig.retina = 3 ) # Load libraries here library(surveydown) library(tidyverse) library(here) ``` # Abstract This study investigates consumer adoption of autonomous delivery technologies—specifically robots and drones for last-mile delivery. The project explores how customers trade off between key delivery attributes: price, speed, environmental sustainability (CO₂ emissions), reliability, and delivery mode (Truck, Robot, Drone). A pilot conjoint survey was conducted among frequent e-commerce shoppers who regularly order small and light packages (under 10 lbs). The results of the conjoint analysis demonstrate that participants made their delivery choices primarily based on price, delivery speed, and environmental impact. Among these, price emerged as the strongest and most statistically significant factor. Consumers showed a clear tendency to avoid higher-priced delivery options, even if other features were attractive, reflecting strong cost sensitivity. Speed was the next most influential factor, with faster deliveries significantly increasing the probability of selection—highlighting the growing consumer expectation for near-instant fulfillment. Emissions also played an important role: lower-carbon delivery options were consistently favored, indicating that environmental consciousness is now part of real consumer decision-making rather than a purely moral statement. By contrast, reliability (within 85–95%) and delivery mode (Truck, Robot, Drone) did not significantly affect choice behavior. This suggests that participants perceive all delivery options as sufficiently reliable and are largely indifferent to whether a human, robot, or drone delivers the package—as long as the service is fast, affordable, and environmentally friendly. In practice, this implies that companies introducing automated delivery technologies should focus their marketing on tangible consumer benefits like cost savings, speed, and sustainability rather than the novelty of the technology itself. The neutrality toward robots and drones indicates readiness for adoption—autonomous delivery is no longer resisted, but to gain market traction, it must deliver clear value in convenience and price. # Introduction The rapid rise of e-commerce has led to an exponential increase in small-package deliveries, making the last-mile segment both critical and costly for logistics companies. Autonomous delivery technologies such as ground robots and drones are being developed to make deliveries faster, cheaper, and more sustainable, but public acceptance remains uncertain. Understanding how consumers value attributes like delivery cost, time, and sustainability is essential before large-scale deployment. <center> <img src="images/drone-delivery.jpg" width=250> </center> This study focuses on evaluating consumer preferences for different delivery modes (Truck, Robot, Drone) and the trade-offs among four major attributes: Price, Speed, CO₂ Emissions, and Reliability. Each attribute was varied at realistic market levels to simulate decision-making scenarios. For example, prices ranged from $2 to $8, speeds from 30 minutes to 24 hours, emissions from 2 lb to 0.001 lb of CO₂ per mile, and reliability from 85% to 95%. A choice-based conjoint (CBC) survey was designed to estimate the relative importance of each attribute. The analysis contributes to the broader policy discussion on the sustainability and feasibility of autonomous last-mile systems, offering insights into consumer willingness to adopt technology-driven delivery methods that could reduce emissions and traffic congestion. # Survey Design ## Eligibility and Target Population The target population comprised frequent online shoppers who regularly receive small or light packages (under 10 lbs). Eligibility criteria ensured relevance to real-world robot delivery applications: - Must order from e-commerce retailers such as Amazon, Walmart, or Target. - Receive at least one package per month. - Typically receive small or light items such as books, electronics, or clothing. - Respondents who only receive large or heavy packages were excluded ## Data Collected The survey collected the following key information: - **Demographics**: age, gender, education level, household income, and residential area. - **Behavioral variables**: online shopping frequency, number of monthly packages, delivery urgency, and typical spending on delivery fees. - **Attitudinal measures**: environmental concern, trust in autonomous delivery, and comfort level with robots or drones ## Educational Material Before choice questions, respondents viewed a brief educational page explaining each delivery attribute in plain language: - **Price**: delivery fee in dollars. - **Speed**: delivery time (e.g., 30 minutes, 1 hour, or 12–24 hours). - **Sustainability (CO₂ emissions)**: pounds of CO₂ per mile emitted during delivery, with examples comparing truck and drone emissions. - **Reliability & Safety**: the likelihood that a package arrives on time and undamaged This step ensured all respondents understood trade-offs before making their choices. ## Conjoint Design Each respondent answered six choice-based conjoint questions, each presenting three delivery options with randomized attribute combinations. No “none” option was included, compelling a realistic trade-off. Attributes and Levels: **Attribute** | **Levels** | **Unit** ----------|--------|-------- **Price** | 2, 5, 8 | $ (dollars) **Speed** | 12-24 hrs, 60 min, 30 min | Time in min & hrs **Sustainability** | 2, 0.05, 0.001 | Pounds of CO₂ per mile **Reliability & Safety** | 85%, 90%, 95% | % successful delivery ## Example Question <center> <img src="images/example.png" width=800> </center> # Data Analysis ## Sample Description The survey sample includes 165 unique respondents who passed the screening criteria and consented, contributing 990 completed choice sets. These respondents also provided answers to demographic and background questions used to describe the sample and explore heterogeneity in delivery preferences. Respondents range from early 20s to over 70 years old, with the age distribution centered in the late 30s to mid‑40s, as shown in the age histogram. Gender is approximately balanced, with slightly more women than men and a very small number of non‑binary respondents. The sample is predominantly white, with smaller shares of Black, Hispanic, Asian, and Native American respondents, as illustrated in the race/ethnicity bar chart. Educational attainment is relatively high overall: most respondents report at least some college, and bachelor’s or associate degrees are the most common categories, with smaller groups having high school only or graduate degrees. Household income is concentrated in the middle ranges, with many respondents between 40,000 and 149,000 USD per year and fewer at the very low or very high ends of the distribution. Package delivery is already a routine activity for this group: most respondents receive packages once a week or several times per week, and only a small minority report rarely receiving packages <center> <img src="figs/sample_description.png" width=800> </center> ## Data Cleaning We began with **215 total survey attempts** in the final deployment. The first quality step was to remove respondents who failed the built-in eligibility screener. In this survey, individuals were screened out if they did not meet the study’s basic criteria—for example, not being an active e-commerce shopper or indicating that they do not order small/light packages. After filtering out screened respondents, **204 responses remained**. This step ensures that the analysis reflects the intended target population of online shoppers whose delivery preferences are meaningful for the study. Next, we restricted the dataset to respondents who **completed all six choice-based conjoint (CBC) tasks**. Anyone missing one or more CBC questions was removed, reducing the dataset from 204 to **201 respondents**. We then applied a standard “flatliner” check to identify respondents who selected the **exact same alternative in every choice task**, a behavior that typically signals low engagement or satisficing rather than thoughtful trade-off evaluation. Respondents exhibiting this pattern were removed, leaving **198 valid respondents** after this step. Finally, we screened for unrealistic completion times, which help identify respondents who may have rushed through the survey without adequately processing the attribute information. We converted total response time into minutes and examined the distribution. Based on this, we excluded respondents who completed the choice-task portion in **under 5 minutes**, a threshold indicating insufficient cognitive effort for evaluating six multi-attribute profiles. After applying the time-quality filter, the final analytic sample consisted of **165 high-quality, engaged respondents**. Thus, out of the original 215 survey attempts, **50 respondents were removed** across screening, completeness, behavioral, and speed checks, ensuring that the remaining dataset was reliable, attentive, and appropriate for estimating the conjoint preference models. <center> <img src="figs/count.png" width=800> </center> ## Modeling ### Simple Logit Model We estimated a Simple Logit model to quantify how each delivery attribute influenced respondents’ choices in the conjoint experiment. $$ u_j = \beta_1 x_j^{\mathrm{price}} + \beta_2 x_j^{\mathrm{speed}} + \beta_3 x_j^{\mathrm{emissions}} + \beta_4 x_j^{\mathrm{reliability}} + \beta_5 \delta_j^{\mathrm{Robot}} + \beta_6 \delta_j^{\mathrm{Drone}} $$ The results for the logit model are summarized below: | parameter | Estimate | Std_Error | Z_value | P_value | Sig | |--------------------|----------|-----------|---------|---------|-----| | price | -0.3564 | 0.0217 | -16.39 | <0.001 | *** | | speed | -0.0349 | 0.0045 | -7.82 | <0.001 | *** | | emissions | -0.5792 | 0.0541 | -10.70 | <0.001 | *** | | reliability | 0.1069 | 0.0119 | 9.00 | <0.001 | *** | | delivery_modeRobot | -0.0053 | 0.1123 | -0.05 | 0.9620 | | | delivery_modeDrone | 0.0765 | 0.1131 | 0.68 | 0.4988 | | The multinomial logit results indicate that several delivery attributes strongly influence consumer choice, while others have minimal impact. **Price**, **emissions**, and **reliability** emerge as the most important drivers of preference. Higher prices and higher CO₂ emissions significantly **reduce** the likelihood that a delivery option is chosen, with emissions showing one of the strongest negative effects, suggesting that respondents place meaningful value on environmentally friendly delivery methods. **Reliability** has a positive and highly significant effect, indicating that consumers prefer options they can depend on. **Speed** also matters, with slower delivery times reducing choice probability, though its effect is smaller than that of price or emissions. In contrast, the **delivery mode** itself—whether a package is delivered by a **robot** or **drone**—shows no statistically significant impact, meaning respondents do not strongly prefer or avoid these newer delivery technologies relative to traditional truck delivery. Overall, consumers prioritize **cost, environmental impact, and performance-related attributes** over the specific technology used to deliver their packages. <center> <img src="figs/model_mnl.png" width=800> </center> ### Mixed Logit | parameter | Estimate | Std_Error | Z_value | P_value | Sig | |-----------|----------|-----------|---------|---------|-----| | price | -0.5109 | 0.0696 | -7.34 | <0.001 | *** | | speed | -0.0559 | 0.0120 | -4.65 | <0.001 | *** | | emissions | -0.8843 | 0.1549 | -5.71 | <0.001 | *** | | reliability | 0.1552 | 0.0277 | 5.61 | <0.001 | *** | | delivery_mode_Drone | 0.2141 | 0.2443 | 0.88 | 0.381 | | | delivery_mode_Robot | -0.2531 | 0.1925 | -1.31 | 0.189 | | | sd_speed | 0.0698 | 0.0270 | 2.58 | 0.009 | ** | | sd_emissions | 0.9519 | 0.2775 | 3.43 | 0.0006 | *** | | sd_reliability | -0.1534 | 0.0692 | -2.22 | 0.027 | * | | sd_delivery_mode_Drone | 0.3441 | 0.9495 | 0.36 | 0.717 | | | sd_delivery_mode_Robot | -1.4459 | 0.5981 | -2.42 | 0.016 | * | The mixed logit results show strong and consistent preferences across respondents: people dislike higher prices, slower delivery, and higher emissions, while valuing greater reliability. Unlike the simple MNL model, the MXL reveals substantial heterogeneity—speed, emissions, reliability, and delivery mode all have significant random-parameter variation, meaning different individuals value these attributes to very different degrees. Delivery mode (Drone or Robot) does not have a significant average effect overall, but the significant variance for the Robot option indicates polarized opinions: some respondents prefer it while others avoid it. The WTP-space model confirms these findings in dollar terms, showing meaningful willingness to pay for improved reliability and reduced emissions. Overall, the mixed logit model provides a better, more realistic representation of individual-level differences in preferences. ### Subgroup Logit Model | Attribute | Group A Mean | A 95% CI Lower | A 95% CI Upper | Group B Mean | B 95% CI Lower | B 95% CI Upper | |---------------------|--------------|----------------|----------------|--------------|----------------|----------------| | speed | -0.098| -0.137| -0.061 | -0.096| -0.132 | -0.063 | | emissions | -1.147 | -1.601 | -0.730 | -1.932 | -2.394| -1.524 | | reliability | 0.214 | 0.117 | 0.316 | 0.357 | 0.265 | 0.459 | | delivery_mode_Drone | 0.265 | -0.657 | 1.179 | 0.209 | -0.654 | 1.072 | | delivery_mode_Robot | 0.199 | -0.723 | 1.098 | -0.106| -0.939 | 0.751 | <center> <img src="figs/subgroup.jpg" width=600> </center> The WTP comparison across income groups reveals meaningful differences in how lower-income (<$50k) and higher-income (≥$50k) respondents value key delivery attributes. Both groups show a positive willingness to pay for enhanced drone delivery and greater reliability, but the effect is notably stronger among higher-income respondents, who exhibit nearly double the WTP for reliability improvements. Emissions reductions also matter to both segments, yet higher-income respondents display a substantially larger negative WTP for higher emissions, indicating a stronger preference for environmentally friendly delivery options. In contrast, speed and robot delivery attributes generate relatively small WTP values in both groups, suggesting these features are less influential in driving choices. Overall, the results indicate that higher-income consumers are more sensitive to environmental quality and service reliability, while lower-income consumers demonstrate more moderate but still positive valuations for premium delivery features. # Results ## WTP | Attribute | Mean WTP ($) | Lower 95% CI | Upper 95% CI | | ------------------ | ------------ | ------------ | ------------ | | **Speed** | -0.098 | -0.124 | -0.073 | | **Emissions** | -1.628 | -1.955 | -1.323 | | **Reliability** | 0.300 | 0.233 | 0.371 | | **Drone delivery** | 0.220 | -0.404 | 0.844 | | **Robot delivery** | -0.012 | -0.619 | 0.607 | The willingness-to-pay (WTP) results reveal clear patterns in the attributes that most strongly influence consumer delivery preferences. Across all plots, increases in delivery speed (longer wait times) and emissions consistently reduce WTP, indicating that consumers prefer faster and environmentally cleaner delivery options and are willing to pay to avoid undesirable levels of each. Reliability emerges as one of the most influential attributes: as reliability increases from 85% to 95%, WTP rises steeply, with wide confidence intervals reflecting substantial heterogeneity in how much consumers value dependable service. In contrast, preferences for delivery mode—drone or robot relative to truck—show small mean WTP values and wide 95% confidence intervals that overlap zero, suggesting that consumers do not strongly prefer one mode over another given current perceptions. Together, these results indicate that **reliability, emissions, and delivery speed are the key drivers of consumer choice**, while delivery mode exerts comparatively weaker influence. <center> <img src="figs/mnl_wtp.png" width=800> </center> ## Market Simulation To evaluate how key delivery attributes influence consumer choice, a set of structured market simulation scenarios was created. Each scenario isolates one attribute—price, emissions, delivery mode, reliability, or speed—while holding the remaining attributes constant. This allows for a clear understanding of how each attribute independently shifts market share among competing alternatives. The baseline attribute levels reflect realistic ranges used in the conjoint survey: - Price: $2, $5, $8 - Speed: 24 hr, 1 hr, 0.5 hr - Emissions: 2 lb, 0.05 lb, 0.001 lb CO₂ - Reliability: 85%, 90%, 95% - Delivery Mode: Truck, Drone, Robot These values were selected because they span low, medium, and high performance levels while remaining consistent with the respondents' choice tasks. The simulations help illustrate competitive positioning when one attribute improves or worsens. <center> <img src="figs/sim_mnl.png" width=300> </center> <center> <img src="figs/sim_multi.png" width=800> </center> **Key Insights from Market Simulations** * **Price is the strongest driver of market share.** Reducing price leads to the largest increases in predicted adoption, making it the most influential attribute in shaping consumer preference. * **Reliability is the most impactful non-price attribute.** Moving from low to high reliability produces a substantial jump in market share, indicating that consumers heavily value trustworthy and consistent delivery performance. * **Speed improvements matter, but only up to a point.** Reducing delivery time from 24 hours to 1 hour significantly boosts preference, but the additional improvement from 1 hour to 0.5 hours provides only a small incremental gain. * **Emissions reductions are valued, but less than reliability or price.** Lowering emissions increases share meaningfully, though the effect is weaker compared to high reliability or lower price. * **Changing delivery mode alone (Drone, Robot, Truck) does not materially shift consumer preference.** When all other attributes are held constant, consumers show similar market shares across modes, suggesting that **performance attributes drive choice more than delivery method**. * **Overall market adoption is maximized by combinations of:** * Lower price * Higher reliability * Faster delivery (major improvements only) * Lower emissions These findings highlight that **improving reliability and optimizing price strategy** offer the greatest opportunities to increase consumer demand, with speed and emissions offering secondary advantages. ## Sensitivity Analysis ### Sensitivity of Market Share to Price * The **market share vs. price** plot shows a smooth, monotonic decline in demand for the drone option (Alternative 2) as price increases from $1 to $20. * At **very low prices (≈ $1–$3)**, predicted share is extremely high (≈ 0.72–0.84), but revenue is still limited because prices are so low. * Around **$5–$6**, market share is still substantial (≈ 0.56 at $5 and ≈ 0.47 at $6) and the **revenue curve peaks**, with total revenue just above $2.7–$3.0 million. * Beyond **$8–$10**, market share falls sharply (below 0.25) and the revenue curve decreases steadily toward zero as price approaches $20. * The grey 95% confidence band around both market share and revenue curves shows that the **revenue maximum is not a single sharp point** but a relatively flat region; prices between roughly **$5 and $6** lie within overlapping confidence intervals and can be viewed as the **best price range** given model uncertainty. ### Tornado Plot – Sensitivity to Other Attributes <center> <img src="figs/tornado_plot.png" width=800> </center> * The tornado plot summarizes how the drone’s market share responds to **±20% changes** in each attribute relative to the baseline (price = $5, speed = 1 hr, emissions = 0.05 kg CO₂, reliability = 90%). * **Reliability** is by far the most influential non-price attribute: * Increasing reliability from 90% to **108%** (a stylized “very high reliability” case) increases predicted share from about **0.56 to 0.90**. * Decreasing reliability to **72%** collapses share to roughly **0.16**. * This very wide bar dominates the tornado chart, indicating that reliability is the **primary design lever** beyond price. * **Price** has the next-largest impact: * Lowering price by 20% (to **$4**) raises share to about **0.64**, while increasing price by 20% (to **$6**) reduces share to about **0.47**. * This confirms the strong price sensitivity already visible in the price–share plot. * **Speed** (1 hr → 0.8 hr or 1.2 hr) and **emissions** (0.05 → 0.04 or 0.06 kg CO₂) have **very small effects** on market share in comparison: * Predicted shares remain clustered around the baseline (≈ 0.56), with only minor deviations, indicating that small ±20% changes in these attributes do not dramatically shift demand in this configuration. ### Design & Pricing Recommendations (with Uncertainty) <center> <img src="figs/rev_price_plot.png" width=600> </center> * **Best price point (revenue-maximizing):** * The revenue curve peaks around **$5–$6**, with overlapping 95% CIs, so this range should be treated as the **optimal pricing band** rather than a single exact value. * Pricing significantly below $5 sacrifices revenue (too much margin lost), while pricing above $6 rapidly erodes both share and revenue. * **Key design decisions to maximize demand:** * **Prioritize reliability**: moving toward very high reliability levels produces the largest gains in market share and should be a central engineering and operations target. * **Use price strategically**: modest under-pricing (e.g., closer to $5 than $6) can secure higher share without sacrificing too much revenue. * **Maintain reasonable speed and low emissions**, but recognize that, within the tested ranges, incremental improvements in these attributes are **less impactful** than reliability and price. * Overall, the sensitivity analysis indicates that **reliability and price are the dominant drivers of market demand**, while speed and emissions play supporting roles. The uncertainty bands around all plots remind us that these estimates are not exact; instead, they define **plausible ranges** within which the true optimal price and attribute levels likely lie. # Final Recommendations and Conclusions ## Overall Competitiveness of the Product The analysis indicates that the proposed delivery product—particularly the drone-based alternative—has strong potential to be competitive in the current market landscape. Across all simulations, drone delivery consistently achieved the highest predicted market share, outperforming both robot and truck delivery modes. The key drivers for its competitiveness are its **low emissions profile** and **high reliability**, which were among the top-valued attributes in the WTP model. The product is therefore well positioned for adoption, provided that pricing remains in a consumer-acceptable range and reliability targets are met. ## Recommended Price Range and Revenue Implications Sensitivity analysis of price demonstrated that market share declines smoothly as price increases, with revenue peaking at approximately **$5–6**, where both adoption and total revenue are maximized. At prices below $4, revenue falls due to margin constraints, while prices above $7 lead to steep declines in market share. Based on the revenue curve and its associated confidence intervals, the optimal price point is **$5**, with reasonable robustness across the confidence band. Estimates remain consistent across simulations, giving moderate confidence in the recommendation, though real-world factors such as retailer partnerships, promotional pricing, or operational costs could shift the optimal point slightly. ## Key Uncertainties Affecting Profitability Despite strong model fit, several uncertainties may influence profitability and competitive success, including: * **Regulatory approval and operational feasibility** for drone or robot delivery, which could delay or limit deployment in certain regions. * **Consumer trust and safety perceptions**, which are not captured fully in the choice experiment but could shape adoption significantly. * **Competitor innovations**, such as ultra-fast truck deliveries or new subscription bundles, which could alter price sensitivity. * **External cost drivers**, such as energy prices, congestion charges, or labor shortages, that may shift the economics of last-mile delivery. These uncertainties imply that price and design recommendations should be updated as technology matures and more real-world performance data becomes available. ## Recommended Actions for Product Design Based on the WTP model, tornado sensitivity analysis, and market simulations: 1. **Prioritize High Reliability:** Reliability showed one of the strongest effects on market share. Increasing reliability from 90% to 108% resulted in a significant market share jump (to ~90%). Even moderate improvements yield noticeable gains. 2. **Maintain Competitive Emissions Performance:** While emissions improvements had slightly smaller marginal effects than reliability, lower emissions levels consistently produced higher adoption. This reinforces sustainability as a market advantage. 3. **Avoid Extremely Fast Delivery Speeds Unless Operationally Efficient:** Speed improves market share, but its effect is smaller relative to price and reliability. Consumers value speed, but not enough to justify major cost trade-offs. 4. **Drone Mode as the Default Offering:** Drone delivery offers the strongest baseline utility and the highest simulated market share. It should be positioned as the flagship offering, with robot delivery reserved for specific contexts (dense urban areas, sidewalks, or building interiors). ## Opportunities for Increasing Demand From the combined analyses, three major opportunities emerge: 1. **Compete Strongly on Reliability:** Reliability is the single most influential attribute in driving preference. Investments in navigation precision, weather handling, and delivery success rates will significantly increase market share and willingness to pay. 2. **Optimize Pricing for Volume and Revenue:** A price point around **$5** balances adoption with strong revenue performance. This price also remains competitive relative to truck-based delivery while differentiating through sustainability and speed. 3. **Leverage Sustainability Messaging:** Emissions reductions meaningfully increase WTP—more than speed improvements. Positioning the service as a *low-carbon delivery option* could expand appeal, especially among environmentally conscious consumers. # Conclusion The combined conjoint results, market simulations, and sensitivity analyses show that a competitively priced, highly reliable, low-emissions drone delivery service has strong market potential. The optimal price range centers around **$5–6**, with reliability improvements offering the greatest leverage for increasing adoption. While uncertainties remain—particularly in regulation, technology maturity, and consumer trust—the model provides robust evidence that the product can achieve meaningful market share and profitability under realistic conditions. Continuous validation with larger and more representative samples is recommended as the product moves closer to implementation. # Limitations - Sample Representatives: Although the final data set contains a sufficient number of completed responses, the sample is not fully representative of the broader population of e-commerce consumers. The survey disproportionately includes younger, student-aged respondents with similar income levels and purchasing patterns, which may not reflect the preferences of high-frequency online shoppers, older users, or individuals in regions with different delivery expectations and infrastructure. - Hypothetical Choice Behavior: Conjoint analysis relies on respondents making choices in a hypothetical environment. While this method reveals trade-offs, real-world behaviors may differ due to brand familiarity, trust in delivery providers, or risk perceptions—factors that were intentionally held constant in this study. Attributes such as drone or robot delivery may also trigger concerns (noise, safety, privacy) that were not measured but could affect true adoption. - Simplified Market Context: The model assumes a market with only three delivery alternatives and holds many contextual variables constant (e.g., delivery fees set by retailers, service availability by region, or time-of-day delivery variations). In reality, consumers face a wider competitive landscape, including same-day premium services, subscription bundles (e.g., Amazon Prime), and retailer-specific policies. - Static and Short-Term Perspective: The study captures preferences at a single point in time. Preferences for sustainability attributes, willingness to pay for speed, or comfort with autonomous systems may evolve as technology matures, regulations change, or users gain more exposure to autonomous delivery options. # Appendix ::: {.sd_page id=welcome} ## Welcome to the Survey Thank you for taking part in our survey. We are graduate students at The George Washington University conducting a study on **"Consumer Adoption of Autonomous Robots for Last-Mile Deliveries"** Your responses are anonymous and will only be used for academic research purposes. There are no right or wrong answers—we are interested in your honest opinions. Click Next to continue. ```{r} sd_next() ``` ::: ::: {.sd_page id=consent} ## Consent This survey is being conducted by students at The George Washing University. We will not be collecting any identifying data, such as your name or address. The whole survey will take approximately 5 to 10 minutes to complete. Your participation is voluntary and you may stop the survey at any time. If you would like to participate, please answer the following questions: ```{r} sd_question( type = 'mc', id = 'consent_age', label = "I am age 18 or older", option = c( 'Yes' = 'yes', 'No' = 'no' ) ) ``` ```{r} sd_question( type = 'mc', id = 'consent_understand', label = "I have read and understand the above information", option = c( 'Yes' = 'yes', 'No' = 'no' ) ) ``` ```{r} sd_next() ``` ::: ::: {.sd_page id=packages} ## Order Packages ```{r} sd_question( type = 'mc', id = 'order_packages', label = "Do you order packages from retailers like Amazon, Walmart etc:", option = c( 'Yes' = 'yes', 'No' = 'no' ) ) ``` ```{r} sd_next() ``` ::: ::: {.sd_page id=packages_info} ```{r} sd_question( type = 'mc_buttons', id = 'frequency', label = "How often do you shop online at retailers like Amazon, Walmart, or Target?", option = c( "Less than Monthly" = "less_monthly", "Monthly once or Twice" = "monthly", "Weekly" = "weekly", "Multiple times per week" = "often" ), direction = "vertical" ) ``` ```{r} sd_question( type = "slider_numeric", id = 'slider_single_val', label = "In a typical month, how many packages (excluding groceries or meals) do you receive from online shopping?", option = seq(0, 10, 1) ) ``` ```{r} sd_next() ``` ::: ::: {.sd_page id=screening} ## Few more Questions ```{r} sd_question( type = 'mc', id = 'screenout', label = "Do you order small packages regularly (under 10 lbs)?", option = c( 'Yes' = 'yes', 'No' = 'no' ) ) ``` ```{r} sd_question( type = "matrix", id = "attitudes_delivery", label = "Please indicate how much you agree or disagree with each statement:", row = c( "I typically need my online orders to arrive within 1 day." = "need_1day", "Environmental impact is an important factor in my delivery choice." = "env_importance", "I would feel comfortable receiving a package delivered by an autonomous robot." = "robot_comfort" ), option = c( "Disagree" = "d", "Neutral" = "n", "Agree" = "a" ) ) ``` ```{r} sd_next() ``` ::: ::: {.sd_page id=context} ## Why We’re Doing This Survey Great work! You are eligible to take our survey. Online shopping has become part of everyday life, and millions of small packages are delivered across cities every day. With so many deliveries happening, companies are now exploring **new ways to deliver packages** faster, cheaper, and more sustainably — including the use of **autonomous delivery robots and drones.** These new technologies could change how we receive our orders, but they also raise important questions: - What do customers value most — cost, speed of delivery, sustainability or reliability? - Would people be comfortable trusting a **robot** or **drone** to bring their package? - How can these services be designed to meet real customer needs? The purpose of this survey is to understand how consumers like you make decisions about different delivery options. Your responses will help researchers identify which delivery features matter most and guide how future autonomous delivery systems are developed and introduced. ```{r} sd_next() ``` ::: ::: {.sd_page id=educational} ## Educational Page Now we'd like you to consider a scenario in which your packages are being delivered by different modes i.e either by **Truck, Robot, Drone**, so you can choose which one do you prefer by comparing their attributes. Before we ask you any questions, let’s learn a little bit more about each of these attributes: ### **Price:** The delivery fee you would pay for a package. ### **Speed:** How quickly your package arrives (examples: within 30 minutes, 1 hour, or 12–24 hours). ### **Sustainability (CO₂ emissions):** The amount of carbon dioxide (CO₂) released per mile during delivery. For reference, a typical delivery truck emits **1 lb of CO₂** per mile. A drone would have to fly **1,000 miles** to emit the same amount of CO₂ as a truck driving just **one mile**. ### **Reliability & Safety:** The chance your package arrives on time, undamaged, and secure. ### **Delivery Modes** We will show 3 different types of delivery modes to choose from: <div style="width: 500px;"> Truck | Robot | Drone | -----|------|------------| <img src="images/truck.png" width=200> | <img src="images/robot.png" width=200> | <img src="images/drone.png" width=200> </div> ```{r} sd_next() ``` ::: ::: {.sd_page id=cbc_practice_page} ## Choice-based Question Practice We'll now begin the choice tasks. On the next few pages we will show you three options of delivery modes and we'll ask you to choose the one you most prefer by comparing all attributes. ```{r} # Define the option vector package_buttons_option <- c("option_1", "option_2", "option_3") # Change the names of each element to display markdown-formatted # text and an embedded image using html names(package_buttons_option)[1] <- " **Option 1** <img src='images/truck.png' width=150> **Delivery Mode**: Truck **Price**: $ 2 **Speed**: 24 hr **Emissions**: 2 lb of CO₂ **Reliability**: 85 % " names(package_buttons_option)[2] <- " **Option 2** <img src='images/robot.png' width=150> **Delivery Mode**: Robot **Price**: $ 5 **Speed**: 1 hr **Emissions**: 0.05 lb of CO₂ **Reliability**: 90 % " names(package_buttons_option)[3] <- " **Option 3** <img src='images/drone.png' width=150> **Delivery Mode**: Drone **Price**: $ 8 **Speed**: 0.5 hr **Emissions**: 0.001 lb of CO₂ **Reliability**: 95 % " sd_question( type = 'mc_buttons', id = 'cbc_practice', label = "Which delivery option would you choose among these?", option = package_buttons_option, ) ``` ```{r} sd_next() ``` ::: ::: {.sd_page id=cbc_intro} ## Conjoint Question Intro Great work! We will now show you 6 sets of choice questions starting on the next page. ```{r} sd_next() ``` ::: ::: {.sd_page id=cbc_q1_page} ## Question 1 ```{r} sd_output("cbc_q1", type = "question") ``` ```{r} sd_next() ``` ::: ::: {.sd_page id=cbc_q2_page} ## Question 2 ```{r} sd_output("cbc_q2", type = "question") ``` ```{r} sd_next() ``` ::: ::: {.sd_page id=cbc_q3_page} ## Question 3 ```{r} sd_output("cbc_q3", type = "question") ``` ```{r} sd_next() ``` ::: ::: {.sd_page id=cbc_q4_page} ## Question 4 ```{r} sd_output("cbc_q4", type = "question") ``` ```{r} sd_next() ``` ::: ::: {.sd_page id=cbc_q5_page} ## Question 5 ```{r} sd_output("cbc_q5", type = "question") ``` ```{r} sd_next() ``` ::: ::: {.sd_page id=cbc_q6_page} ## Question 6 ```{r} sd_output("cbc_q6", type = "question") ``` ```{r} sd_next() ``` ::: ::: {.sd_page id=demographics} ## Demographics Page ### Nice job! We're almost done! We'd just like to ask just a few more questions about you which we will only use for analyzing our survey data. ::: {style="width:500px; margin: 0 auto;"} ```{r} years <- as.character(2003:1920) names(years) <- years years <- c("Prefer not to say" = "prefer_not_say", years) sd_question( type = 'select', id = 'year_of_birth', label = "(1) In what year were you born?", option = years ) ``` ```{r} genders <- c( "Male" = "male", "Female" = "female", "Non-binary" = "non_binary", "Trans male/trans man" = "trans_male", "Trans female/trans woman" = "trans_female", "Prefer not to say" = "prefer_not_to_say" ) sd_question( type = 'select', id = 'gender', label = "(2) What is your current gender identity?", option = genders ) ``` ```{r} races <- c( "White (Not of Hispanic or Latino origin)" = "white", "African American or Black" = "black", "Asian" = "asian", "Hispanic or Latino" = "hispanic", "American Indian or Alaska Native" = "native_american", "Native Hawaiian or Other Pacific Islander" = "pacific_islander", "Prefer not to say" = "prefer_not_to_say" ) sd_question( type = 'select', id = 'race', label = "(3) I identify my race as (select all that apply):", option = races ) ``` ```{r} educations <- c( "Less than high school" = "less_than_high_school", "High school" = "high_school", "Some college" = "some_college", "Associate's degree" = "associate_degree", "Bachelor's degree" = "bachelor_degree", "Master's degree" = "master_degree", "Doctoral degree" = "doctoral_degree", "Prefer not to say" = "prefer_not_to_say" ) sd_question( type = 'select', id = 'education', label = "(4) What is the highest degree or level of school you have completed? If currently enrolled, please use the highest degree received.", option = educations ) ``` ```{r} incomes <- c( "Less than $10,000" = "less_than_10k", "$10,000 to $14,999" = "10k_to_14k", "$15,000 to $19,999" = "15k_to_19k", "$20,000 to $24,999" = "20k_to_24k", "$25,000 to $29,999" = "25k_to_29k", "$30,000 to $34,999" = "30k_to_34k", "$35,000 to $39,999" = "35k_to_39k", "$40,000 to $49,999" = "40k_to_49k", "$50,000 to $74,999" = "50k_to_74k", "$75,000 to $99,999" = "75k_to_99k", "$100,000 to $149,999" = "100k_to_149k", "$150,000 to $199,999" = "150k_to_199k", "$200,000 or more" = "200k_or_more", "Prefer not to say" = "prefer_not_to_say" ) sd_question( type = 'select', id = 'income', label = "(5) What is your annual household income (from all sources) before taxes and other deductions from pay?", option = incomes ) ``` ```{r} sd_question( type = "textarea", id = "feedback", label = "Please let us know if you have any other thoughts or feedback on this survey. Your feedback will help us make future improvements :)" ) ``` ```{r} sd_next() ``` ::: ::: ::: {.sd_page id=end_consent} ## End Page The survey is now finished. You may close the window. ```{r} sd_close() ``` :::